VibeMathedMath problems solved with AI

Tao's question on the relations of a Lie subalgebra

Let g\mathfrak g be the real Lie algebra generated by Di,EiD_i,E_i (i=1,2,3,…i=1,2,3,\dots) subject to [Di,Dj]=[Ei,Ej]=0[D_i,D_j]=[E_i,E_j]=0 and [Di,Ej]=[Dj,Ei][D_i,E_j]=[D_j,E_i]. Let h\mathfrak h be the subalgebra generated by the [Di,Ej][D_i,E_j] and the EiE_i. It satisfies [Di,Ej]=[Dj,Ei][D_i,E_j]=[D_j,E_i], [Ei,Ej]=0[E_i,E_j]=0 and, by the Jacobi identity, [Ei,[Dj,Ek]]=[Ek,[Dj,Ei]][E_i,[D_j,E_k]]=[E_k,[D_j,E_i]]. Are these a generating set of relations for h\mathfrak h, or are there further relations among the [Di,Ej][D_i,E_j] and EiE_i? In a comment YCor also asked whether the [Di,Ej][D_i,E_j], i≤ji\le j, freely generate a Lie subalgebra of g\mathfrak g.

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI-assisted
Method
Argument
Field
Lie algebras; presentations
Posed by
Terry Tao, MathOverflow question 464808 (24 February 2024)
Year posed
2024
Years open
2y
Solved
2026-11-16
Model
OpenAI Codex; Anthropic Claude (versions not recorded)
Vendor
OpenAI; Anthropic
Collaborators
Qiyuan Ma
Verification
Unreviewed
Publication
Preprint
Significance
10 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Theorem 1.1: yes. The three families of relations, with the Lie algebra axioms, generate all relations of h\mathfrak h, over any field of characteristic 0. The proof polarises to multilinear components, presents the ideal generated by the EiE_i, bounds an associated graded algebra for an 'excess' filtration, and closes with a dimension count from the Hilbert series of g\mathfrak g (Eulerian numbers). By-products: h\mathfrak h is a semidirect product of a free Lie algebra and an abelian one, and the [Di,Ej][D_i,E_j], i≤ji\le j, freely generate a free Lie subalgebra, answering YCor's question. The paper says the Hilbert series is equivalent to known results on Losev-Manin spaces (Losev-Manin 2000; Dotsenko-Shadrin-Tamaroff 2023) and claims as new only the answer and the freeness statement, which its literature search did not find elsewhere.

What the AI did

From the paper's acknowledgements and the Hexagon provenance statement: AI assistance (Codex from OpenAI and Claude from Anthropic, versions not recorded) was used in the literature search, in designing and running the computational experiments, in writing the accompanying verification code, and in drafting the manuscript. Hexagon labels the paper 'Primarily AI-generated text', human understanding 'some parts'; the author reports that some parts, but not all, have been checked and understood, without saying which, and takes sole responsibility. AI is not an author.

Verification

Read by this site on 19 November 2026: the Hexagon record and PDF (hexagon 2611.00542v1, 17 pages) and Tao's MathOverflow question with all its comments, fetched through the Stack Exchange API. The statement matches the question as posed (the paper works over any field of characteristic 0, the question over the reals), and the question had no answers. A sanity computation written here, independent of the author's code, built the multilinear components of the enveloping algebras by linear algebra and found the map from the presented algebra injective in every multilinear degree n≤5n\le 5 (dimensions 2, 5, 18, 77 on both sides), and the dimensions of U(g)U(\mathfrak g) there (5, 16, 65, 326) consistent with the paper's (n−1)!(n-1)! count for g\mathfrak g. That supports the theorem in low degree; it does not check the proof. The 17-page argument was not refereed, the author's code archive was not located or run, and by the Hexagon label the author has checked only some parts. No expert reaction was found. Candidate until someone independent reads the proof in public.

Sources

Submitted by RustyDingo867 on

Changelog2 changes

Discussion